1. Field of the Invention
The present invention relates to an image coding system for coding an image signal with high efficiency.
2. Description of the Prior Art
As is known in the art, means for eliminating redundant components included in an image signal is used for coding an image signal. A typical approach to image coding is the transform coding method wherein an image is divided into blocks, an orthogonal transform is carried out for each of the blocks, and the transform coefficients are encoded.
In the case of television signals such as an NTSC signal, interlaced scanning is used whereby an image signal of one frame is scanned twice, once in the odd field and once in the even field. The two fields scan different but complementary spaces of an image. The fields have image information at different times but there is a relatively strong correlation therebetween because the scanned lines of the two fields are alternate and adjacent. There is a technique in which coding is carried out after combining the fields and dividing them into blocks when coding an image signal produced by the interlaced scanning.
FIG. 1 is a block diagram showing the structure of an embodiment of "High Efficiency Image Coding System" described in the Japanese Patent Public Disclosure No. 1688/1991. In FIG. 1, the coding system includes a non-interlacing section 1, a motion detecting section 2, a non-interlace blocking section 3, an individual field blocking section 4, an orthogonal transform section 5, a quantizing section 6 for quantizing a conversion coefficient at the output of the orthogonal transform section 5, and coding section 7.
In operation, a series of input image signals 100, which are produced by the interlaced scanning method and applied to each field, are converted to a non-interlaced signal 101 in the non-interlacing section 1 as indicated in FIG. 2(C). As shown, the pixels belonging to the odd field and the pixels belonging to the even field appear alternately in every other line.
When an object is stationary and the correlation between adjacent lines is high, it is effective to use a non-interlaced signal and to code the image signal in a block including components from both fields. FIG. 3(A) shows an example of such a condition. On the other hand, when an object is moving, the correlation between adjacent lines is lowered and it is considered to be effective to execute the coding in units of individual fields. This is because a non-interlaced signal is used for the moving object results in discontinuation as shown in FIG. 3(B), causing a power to be generated in high frequency coefficients during the transform coding. In this case, the blocking as indicated in FIG. 3(C) is adequate.
Thus, the motion detector 2 detects the motion of an object and changes the operation when the object is detected as being stationary by a signal 103 indicating motion, to conduct the blocking shown in FIG. 3(A) (hereinafter, this arrangement of FIG. 3(A) is called the non-interlace blocking) in the non-interlace blocking circuit 3. If the object is detected to be moving, the motion detector 2 changes the operation to conduct the blocking shown in FIG. 3(C) (hereinafter, this arrangement of FIG. 3(C) is called the individual field blocking) in the individual field blocking circuit 4.
The blocks obtained by changing the blocking as explained above are subjected to the discrete cosine transformation (DCT) in the orthogonal transform section 5. The transform coefficients obtained as described above are quantized in the quantizing section 6, and a variable length code is assigned in the coding section 7 in accordance with the occurrence probability of respective events.
Since a conventional image coding system has been structured as described above, it has been difficult to realize the blocking utilizing the correlation between fields when an object is moving. Moreover, such a system has not utilized the property of different intensities in power distribution of the coefficients after conversion caused by the difference in arrangement of pixels within the block. In addition, there is the difference in power between the stationary blocks and moving blocks. The moving blocks having a high signal power which has not been utilized.
FIG. 4 is a block diagram of another conventional interframe predictive coding system described, for example, in the transactions on the 3rd HDTV International Work Shop, "A Study on HDTV Signal Coding with Motion Adaptive Noise Reduction" (Vol 3, 1989). In FIG. 4, this system comprises a frame memory 21, a motion detecting section 22. a subtracter 23, a coding section 24, a local decoding section 25, an adder 26 and a multiplexing section 27. Although omitted in this figure, the encoded date is decoded at a receiving side in order to reproduce the transmitted signal.
In operation, the motion of an object between the current field and the field of the same type of the preceding frame is detected block by block, the block consisting of a plurality of pixels of an input image signal 201 which is provided by the interlaced scanning method and formed of frames, each frame having both odd and even fields. The motion between odd fields is detected in the motion detecting section 22 by searching the block which has the most distinctive resemblance to the currently processing block among the already encoded blocks 202, adjacent to the position corresponding to the currently processing block in the odd fields stored within the frame memory 91. The degree of resemblance is evaluated by using an absolute sum of differential values or a square sum of differential values of the corresponding pixels in both blocks. The amount of motion in both horizontal and vertical directions between the current block and the block determined to be the most similar is provided as a motion vector 203. The frame memory 21 outputs a motion compensated prediction signal 204 corresponding to this motion vector 203.
A prediction error signal 205 obtained in the subtracter 23 by subtracting the motion compensated prediction signal 204 from the input signal 201 :is applied to the coding circuit 24 in which the spatial redundancy is removed. Since low frequency components of an image signal generally occupy a greater part of the power thereof information can be compressed by quantizing high power portions with a large number of bits and quantizing low power portions with a small number of bits. According to an example of this information compression method, the frequency conversion is carried out for an 8.times.8 pixels block by conducting an orthogonal transform such as a discrete cosine transform to scalar-quantize the transform coefficients. The scalar-quantized coding data 206 is sent to the local decoding section 25 and to the multiplexing section 27. The multiplexing section 27 conducts multiplexing and encoding for the coding data 206 and the motion vector 203 to output these signals to a transmission line 209.
Meanwhile, the local decoding circuit 25 executes the inverse operation of the operation in the coding section 24, namely the inverse scalar quantization and inverse orthogonal transform to obtain a decoded error signal 207. The motion compensated prediction signal 204 is added to the decoded error signal 207 in the adder 26 and stored in the frame memory 21 to detect motion of the odd field of the next frame.
In addition, the motion of the even fields of the input image signal 201 with respect to the already encoded field of the frame memory 21 is also detected for the coding of the motion compensated prediction error signal. As described above, in the conventional interframe predictive coding system, redundancy with respect to time included in moving image signals is removed by the motion compensated prediction coding and redundancy with respect to space is removed by the orthogonal transform.
Since the conventional interframe predictive coding system is structured to individually encode both the odd field and even field by predicting the current (present) odd field from the odd field of the already encoded frame and predicting the current even field from the even field of the already encoded frame, the encoding efficieny is low because the spatial correlation existing between the continuous fields, produced by the interlaced scanning method, is not used.